Classifications

• • G—PHYSICS
  • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
  • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
  • G16B25/00—ICT specially adapted for hybridisation; ICT specially adapted for gene or protein expression
• • B—PERFORMING OPERATIONS; TRANSPORTING
  • B82—NANOTECHNOLOGY
  • B82Y—SPECIFIC USES OR APPLICATIONS OF NANOSTRUCTURES; MEASUREMENT OR ANALYSIS OF NANOSTRUCTURES; MANUFACTURE OR TREATMENT OF NANOSTRUCTURES
  • B82Y10/00—Nanotechnology for information processing, storage or transmission, e.g. quantum computing or single electron logic
• • B—PERFORMING OPERATIONS; TRANSPORTING
  • B82—NANOTECHNOLOGY
  • B82Y—SPECIFIC USES OR APPLICATIONS OF NANOSTRUCTURES; MEASUREMENT OR ANALYSIS OF NANOSTRUCTURES; MANUFACTURE OR TREATMENT OF NANOSTRUCTURES
  • B82Y5/00—Nanobiotechnology or nanomedicine, e.g. protein engineering or drug delivery
• • G—PHYSICS
  • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
  • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
  • G16B25/00—ICT specially adapted for hybridisation; ICT specially adapted for gene or protein expression
  • G16B25/10—Gene or protein expression profiling; Expression-ratio estimation or normalisation
• • G—PHYSICS
  • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
  • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
  • G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding
  • G16B40/10—Signal processing, e.g. from mass spectrometry [MS] or from PCR
• • G—PHYSICS
  • G16—INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR SPECIFIC APPLICATION FIELDS
  • G16B—BIOINFORMATICS, i.e. INFORMATION AND COMMUNICATION TECHNOLOGY [ICT] SPECIALLY ADAPTED FOR GENETIC OR PROTEIN-RELATED DATA PROCESSING IN COMPUTATIONAL MOLECULAR BIOLOGY
  • G16B40/00—ICT specially adapted for biostatistics; ICT specially adapted for bioinformatics-related machine learning or data mining, e.g. knowledge discovery or pattern finding

Definitions

• the invention generally relates to techniques for analyzing biological samples such as DNA, RNA, or protein samples and in particular to techniques for analyzing the output patterns of hybridized biochip microarrays.
• DNA-based analysis may be used either as an in-vitro or as an in-vivo control mechanism to monitor progression of disease, assess effectiveness of therapy or be used to design dosage formulations. DNA-based analysis is used verify the presence or absence of expressed genes and polymorphisms.
• a DNA microarray includes a rectangular array of immobilized single stranded DNA fragments. Each element within the array includes few tens to millions of copies of identical single stranded strips of DNA containing specific sequences of nucleotide bases. Identical or different fragments of DNA may be provided at each different element of the array. In other words, location (1,1) contains a different single stranded fragment of DNA than location (1,2) which also differs from location (1,3) etc. Certain biochip designs may replicate the nucleotide sequence in multiple cells.
• DNA-based microarrays deploy chemiluminiscence, fluorescence or electrical phenomenology to achieve the analysis.
• a target DNA sample to be analyzed is first separated into individual single stranded sequences and fragmented. Each sequence being tagged with a fluorescent marker molecule.
• the fragments are applied to the microarray where each fragment binds only with complementary DNA fragments already embedded on the microarray. Fragments which do not match any of the elements of the microarray simply do not bind at any of the sites of the microarray and are discarded during subsequent fluidic reactions. Thus, only those microarray locations containing fragments that bind complementary sequences within the target DNA sample will receive the fluorescent molecules.
• a fluorescent light source is then applied to the microarray to generate a fluorescent image identifying which elements of the microarray bind to the patient DNA sample and which do not.
• the image is then analyzed to determine which specific DNA fragments were contained within the original sample and to determine therefrom whether particular diseases, mutations or other conditions are present in the patient sample.
• a particular element of the microarray may be exposed to fragments of DNA representative of a particular type of cancer. If that element of the array fluoresces under fluorescent illumination, then the DNA of the sample contains the DNA sequence representative of that particular type of cancer. Hence, a conclusion can be drawn that the patient providing the sample either already has that particular type of cancer or is perhaps predisposed towards that cancer. As can be appreciated, by providing a wide variety of known DNA fragments on the microarray, the resulting fluorescent image can be analyzed to identify a wide range of conditions.
• the step of analyzing the fluorescent pattern to determine the nature of any conditions characterized by the DNA is expensive, time consuming, and somewhat unreliable for all but a few particular conditions or diseases.
• One major problem with many conventional techniques is that the techniques have poor repeatability. Hence, if the same sample is analyzed twice using two different chips, different results are often obtained. Also, the results may vary from lab to lab. Consistent results are achieved only when the target sample has high concentrations of oligonucleotides of interest. Also, skilled technicians are required to prepare DNA samples, implement the hybridization protocol, and analyze the DNA microarray output possibly resulting in high costs.
• fluorescently labeled primers are prepared for flanking loci of genes of interest within the DNA sample.
• the primers are applied to the DNA sample such that the fluorescently labeled primers flank genes of interest.
• the DNA sample is fragmented at the locations where the fluorescently labeled primers are attached to the genes of interest to thereby produce a set of DNA fragments, also called "oligonucleotides" for applying to the DNA microarray.
• DNA microarrays there are two types: passive hybridization microarrays and active hybridization microarrays.
• passive hybridization oligonucleotides characterizing the DNA sample are simply applied to the DNA microarray where they passively attach to complementary DNA fragments embedded on the array.
• active hybridization the DNA array is configured to externally enhance the interaction between the fragments of the DNA samples and the fragments embedded on the microarray using, for example, electronic techniques.
• FIG. 1 both passive hybridization and active hybridization steps are illustrated in parallel. It should be understood that, currently for any particular microarray, either the passive hybridization or the active hybridization steps, but not both, are typically employed.
• a DNA microarray chip is prefabricated with oligonucleotides of interest embedded or otherwise attached to particular elements within the microarray.
• the oligonucleotides of the DNA sample generated at step 102 are applied to the microarray. Oligonucleotides within the sample that match any of the oligonucleotides embedded on the microarray passively bind with the oligonucleotides of the array while retaining their fluorescently labeled primers such that only those locations in the microarray having corresponding oligonucleotides within the sample receive the primers.
• each individual nucleotide base within the oligonucleotide sequence (with lengths ranging from 5 to 25 base pairs) can bond with up to four different nucleotides within the microarray, but only one oligonucleotide represents an exact match. When illuminated with fluorescent light, the exact matches fluoresces most effectively and the non-exact matches fluoresce considerably less or not at all.
• the DNA microarray with the sample loaded thereon is placed within a fluidics station provided with chemicals to facilitate the hybridization reaction, i.e., the chemicals facilitate the bonding of the oligonucleotide sample with corresponding oligonucleotides within the microarray.
• the microarray is illuminated under fluorescent light, perhaps generated using an ion-argon laser, and the resulting fluorescent pattern is digitized and recorded. Alternately, a photograph of the fluorescent pattern may be taken, developed, then scanned into a computer to provide a digital representation of the fluorescent pattern.
• the digitized pattern is processed using dedicated software programs which operate to focus the digital pattern and to subsequently quantize the pattern to yield a fluorescent intensity value for each array within the microarray pattern.
• the resulting focused array pattern is processed using additional software programs which compute an average intensity value at each array location and provides for necessary normalization, color compensation and scaling.
• a digitized fluorescent pattern has been produced identifying locations within the microarray wherein oligonucleotides from the DNA sample have bonded. This fluorescent pattern is referred to herein as a "dot spectrogram".
• a DNA microarray is prefabricated for active hybridization at step 116.
• the DNA sample is applied to the active array and electronic signals are transmitted into the array to help facilitate bonding between the oligonucleotides of the sample and the oligonucleotides of the array.
• the microarray is then placed within a fluidics station which further facilitates the bonding.
• an electronic or fluorescent readout is generated from the microarray.
• step 124 the electronic output is processed to generate a dot spectrogram similar or identical to the dot spectrogram generated using the optical readout technique of steps 110-114.
• the result is a dot spectrogram representative of oligonucleotides present within the DNA sample.
• some conventional passive hybridization DNA microarrays provide electronic output and some active hybridization microelectronic arrays provide optical readout.
• the output of step 108 is processed in accordance with steps 122 and 124.
• the output of step 120 is processed in accordance with steps 110-114. Again, the final results are substantially the same, i.e., a dot spectrogram.
• step 126 the dot spectrogram is analyzed using clustering software to generate a gene array amplitude readout pattern representative of mutations of interest within the target DNA sample.
• step 126 operates to correlate oligonucleotides represented by the dot spectrogram with corresponding DNA mutations.
• step 128, the resulting digital representation of the mutations of interest are processed using mapping software to determine whether the mutations are representative of particular diagnostic conditions, such as certain diseases or conditions.
• step 128 operates to perform a mutation-to-diagnostic analyses.
• step 130 the diagnostic conditions detected using step 128 are evaluated to further determine whether or not the diagnostic, if any, can properly be based upon the DNA sample.
• Classical methods such as probabilistic estimator such as minimum a posteriori (MAP) estimator, maximum likelihood estimator (MLE) or inferencing mechanism may be used to render a diagnostic assessment.
• MAP minimum a posteriori
• MLE maximum likelihood estimator
• the invention is directed primarily to providing a sequence of steps for replacing steps 114-130 of FIG. 1.
• a method for analyzing an output pattern of a biochip to identify mutations, if any, present in a biological sample applied to the biochip.
• a resonance pattern is generated which is representative of resonances between a stimulus pattern associated with a set of known mutations and the output pattern of the biochip.
• the resonance pattern is interpreted to yield a set of confirmed mutations by comparing resonances found therein with predetermined resonances expected for the selected set of mutations.
• the biological sample is a DNA sample and the output pattern being analyzed is a quantized dot spectrogram generated by a hybridized oligonucleotide microarray.
• the resonance pattern is generated by iteratively processing the dot spectrogram by performing a convergent reverberation to yield a resonance pattern representative of resonances between a predetermined set of selected Quantum Expressor Functions and the dot spectrogram until a predetermined degree of convergence is achieved between the resonances found in the resonance pattern and resonances expected for the set of mutations.
• the resonance pattern is analyzed to yield a set of confirmed mutations by mapping the confirmed mutations to known diseases or diagnostic conditions of interest, associated with the pre-selected set of known mutations to identify diseases, if any, indicated by the DNA sample.
• a diagnostic confirmation is then made by taking the identified diseases and solving in reverse for the associated Quantum Expressor Functions and then comparing those Quantum Expressor Functions with ones expected for the mutations associated with the identified disease to verify correspondence. If no correspondence is found, a new sub-set of known mutations are selected and the steps are repeated to determine whether any of the new set of mutations are present in the sample.
• the set of nonlinear Quantum Expressor Functions are generated are follows.
• a set of mutation signatures representative of the pre-selected set of known mutations is input.
• a representation of a microarray oligonucleotide pattern layout for the microarray, from which the dot spectrogram is generated, is also input.
• a set of resonant interaction parameters are generated which are representative of mutation pattern interactions between elements of the microarray including interactions from a group including element-to- element interactions, element-to-ensemble interactions, ensemble-to-element interactions, and ensemble-to-ensemble interactions.
• the set of nonlinear Quantum Expressor Functions are generated from the set of resonant interaction patterns by matching selected harmonics of the power spectral density (PSD) amplitude of a coded mutation signature, corresponding to the preselected mutation set of interest, to that of a pre-determined quantum-mechanical Hamiltonian system so that stochastic and deterministic time scales match, and the time scales couple back to noise statistics and degree of asymmetry.
• PSD power spectral density
• the dot spectrogram is differentially enhanced prior to the generation of the resonance pattern by refocusing the dot spectrogram to yield a re-focused dot spectrogram; cross-correlating the re-focused dot spectrogram; applying a local maxima filter to the correlated re-focused dot spectrogram to yield a maximized dot spectrogram; re-scaling the maximized dot spectrogram to yield a uniformly re-scaled dot spectrogram; and then re-scaling the uniformly re-scaled dot spectrogram to amplifying local boundaries therein to yield a globally re- scaled dot spectrogram.
• a method of generating a set of nonlinear Quantum Expressor Functions includes the steps of inputting a set of mutation signatures representative of the pre-selected set of known mutations and inputting a representation of a biochip layout.
• the method also includes the steps of generating a set of resonant interaction parameters representative of mutation pattern interactions between elements of the microarray including interactions from a group including element-to-element interactions, element-to-ensemble interactions, ensemble-to-element interactions, and ensemble-to-ensemble interactions and generating the set of nonlinear Quantum Expressor Functions from the set of resonant interaction patterns.
• principles of the invention are applicable to the analysis of various arrayed biomolecular, ionic, bioelectronic, biochemical, optoelectronic, radio frequency (RF) and electronic microdevices.
• Principles of the invention are particularly applicable to mutation expression analysis at ultra-low concentrations using ultra-high density passive and/or active hybridization DNA-based microarrays.
• Techniques implemented in accordance with the invention are generally independent of the physical method employed to accumulate initial amplitude information from the bio-chip array, such as fluorescence labeling, charge clustering, phase shift integration and tracer imaging.
• principles of the invention are applicable to optical, optoelectronic, and electronic readout of hybridization amplitude patterns.
• principles of the invention are applicable to molecular expression analysis at all levels of abstraction: namely DNA expression analysis, RNA expression analysis, protein interactions and protein - DNA interactions for medical diagnosis at the molecular level. Apparatus embodiments are also provided.
• Fig. 1 is a flow chart illustrating conventional passive and active hybridization DNA microarray analysis techniques.
• Fig. 2 is a flow chart illustrating an exemplary method for analyzing the output of a hybridized DNA microarray in accordance with the invention.
• Fig.3 graphically illustrates the method of Fig.2.
• Fig.4 is a flow chart illustrating an exemplary method for generating Quantum Expressor Functions for use with the method of Fig. 2.
• Fig.5 is a flow chart illustrating an exemplary method for preconditioning the output of a hybridized DNA microarray for use with the method of Fig.2.
• DNA samples but principles of the invention apply to the analysis of other protein-based samples or to other types of output patterns as well.
• the exemplary method exploits, among other features: (a) a novel representation, interpretation and mathematical model for the immobilized oligonucleotide hybridization patterns, represented via a dot spectrogram; (b) a new "active" biomolecular target detection and discrimination method based on quantum resonance interferometry, and (c) a new spatial hashing function that yields accurate diagnostic assessment.
• the exemplary method exploits a fundamentally different computational paradigm for mutation expression detection in pre-enhanced dot spectrogram realizations.
• the method is an innovative modification to dynamically arrayed quantum stochastic resonance (QSR) for discrete system analysis.
• QSR quantum stochastic resonance
• the arraying strategy is a function of the expression pathway of interest.
• the method depends on the molecular diagnostic spectrum being addressed. Coupled quantum resonators are employed to significantly enhance signal-to-noise (SNR) performance and fuse multiple synthetic renormalized dot spectrogram realizations to better detect prespecified biomolecular expression patterns.
• SNR signal-to-noise
• the exemplary method exploits an enhancement in previous extensions to classical stochastic resonance (SR) and array enhanced SR (AESR) in signal processing and sensor data analysis.
• Stochastic resonance is a phenomenon wherein the response to a sensor, modeled in terms of a bistable nonlinear dynamical system, is enhanced by applying a random noise element and a periodic sinusoidal forcing function.
• SR occurs when the SNR passes through a maximum as the noise level is increased.
• an important aspect of the exemplary method involves the coupling of transformed and preconditioned discrete microarray outputs to a mathematical model for a quantum-mechanical dynamical system with specific properties. When driven in a particular manner, the coupled system exhibits a nonlinear response that corresponds to detection of phenomena of interest.
• the method exploits modulation of observables from a "base” (canonical continuous dynamical system), so that a selected set of spectral properties match a similar selected spectral properties of a discrete spatial tessellation substructure from an amplitude spectrogram derived from bioelectronic observables.
• the method further exploits the concept of convolving a discrete spatial system (derived from base mutants of interest) with a continuous asymmetric temporal system to derive a spatiotemporal input to further convolve with another discrete spatial projection (of an inherently partially stabilized spatiotemporal system).
• exemplary biomolecular detection method selection of a basis system; (ii) generation of designer Quantum Expressor Function (QEF) for coupling with the substrate to be analyzed; (iii) generation of a Hamiltonian to describe relaxation dynamics of the coupled system; (iv) modulation of resonance parameters to enforce early resonance; (v) and exploitation of resonance suppressors to verify detection.
• QEF Quantum Expressor Function
• a set of mutations of interest is selected.
• the mutations may be mutations relevant to cancer, AIDS, or other diseases or medical conditions.
• preconditioner transforms are generated based upon the selected set of mutations.
• the preconditioner transforms are provided to convert mutation nucleotide sequences into expected amplitude patterns in the prespecified microarray representation, given a particular biochip layout.
• Quantum Expressor Functions are generated based upon the Hamiltonian of a pre-selected basis system.
• the Quantum Expressor Functions are designed to couple the Hamiltonian for the selected basis system to a predetermined DNA microarray configuration to permit a resonance interaction involving the output of the DNA microarray.
• Resonance stimulus is generated, at step 204, using the Quantum Expressor functions.
• any number of output patterns from the DNA microarray may be processed using the Quantum Expressor Functions to identify whether any of the mutations of the pre-selected set of mutations are found therein.
• Quantum Expressor Functions are pre-generated for a large set of mutations and for a large set of DNA microarray patterns such that, for each new DNA microarray output pattern from each new patient sample, the presence of any of the mutations can be quickly identified using the predetermined set of Quantum Expressor Functions.
• an output pattern (referred to herein as a Dot Spectrogram) is generated using a DNA microarray for which Quantum Expressor Functions have already been generated.
• the dot spectrogram is preconditioned to yield a dot spectrogram tesselation (DST) to permit exploitation of a resonance interaction between the dot spectrogram and the Quantum Expressor Functions.
• DST dot spectrogram tesselation
• step 212 a resulting resonance pattern is processed at step 212 to identify any mutations represented thereby.
• step 212 is rendered trivial by virtue of the aforementioned resonant interaction which is based upon Quantum Expressor Function already correlated with the pre-selected mutations. Hence, no complicated analysis is required to interpret the resonance pattern to identify the mutations.
• step 214 the mutations are mapped to corresponding diseases and conditions to thereby identify any diseases or conditions that the patient providing the sample being analyzed is afflicted with. Again, this is a fairly trivial step.
• diagnostic confirmation is preformed to verify that the diseases or conditions are present in the sample. This is achieved by starting with the found diseases or conditions and then performing the steps of the method in reverse.
• FIG. 3 graphically illustrates the operation of the method of FIG.2 whereby a Quantum Expressor Function 300 is generated based on a mutation set 302.
• a dot spectrogram 304 is applied to the Quantum Expressor Function via an interferometric resonance interaction 306 yielding a resonance pattern (not shown) from which mutations signatures 308 (representative of mutations present in the sample from which the dot spectrogram was generated) may be identified by comparison with mutation set 302.
• the mutation set of interest generated at step 200 is selected by identifying oligonucleotides representative of the ⁇ Z ⁇ mutations of interest.
• length
• an oligonucleotide table is generated which contains the oligonucleotide sequences associated with each mutation of interest identified by row and column location (i,j).
• the oligonucleotide table is provided for subsequent use at step 212 to map locations within the dot spectrogram wherein resonance occurs at step 210 to oligonucleotides such that mutations present in a sample being analyzed are easily identified.
• a mutation table is generated which contains the diseases associated with each mutation of interest.
• the mutation table is provided for subsequent use at step 214 to map mutations identified at step 212 to specific diseases or other medical conditions such that the diseases can be easily identified.
• the DNA microarray is an N by M DNA chip array wherein an element of the array is referred to herein as an "oxel": o(ij).
• PEBC pre-hybridization microarray
• N M PEBC X X (i,j) , where N and M refer to the linear (row and column)
• SVD singular value decomposition
• QEF Quantum Expressor Functions
• h(ij) An element of the dot spectrogram is referred to herein as a hixel: h(ij).
• CSR complete spatial randomness
• DST algorithm on preconditioned hixel array, such as variants of Dirichlet tessellation would further reduce it to ⁇ 0( 10) with no coding. Consequently, this method must be applied to ⁇ O( 10) ensembles.
• the ensemble basis system is degenerate, i.e., it has (i) a bounded intensity function; (ii) a bounded radial distribution function, (iii) and is anisotropic.
• the Quantum Expressor Functions (QEF's) generated at step 202 are based upon the DNA chip used to generate the dot spectrogram and based upon the mutation set of interest. More specifically, as shown in FIG. 4, the QEF is generated based upon the spin Boson basis system by first calculating the Hamiltonian for the system at step 402 then, at step 404, calculating harmonic amplitudes
• OF order function
• QSR quantum stochastic resonance
• Classical QSR is an archetypal example of a phenomenon where quantum noise is exploited to drive order in a quantum-mechanical system, as opposed to Gaussian noise in a classical system.
• the appearance of a resonance requires an asymmetry in the energies of the two states.
• a rate equation can be constructed for the system, such that the dynamics can be characterized in terms of transition rates ⁇ + and ⁇ - between the two asymmetric quantum superposition states, and when the drive frequency and the interwell transition rates are much slower than the intrawell relaxation rates.
• SNR signal to noise ration
• QSR occurs for an asymmetric well, but not for a symmetric energy well.
• ⁇ denotes the asymmetric energy
• ⁇ is the tunneling matrix element
• ⁇ is the Pauli spin matrices
• ⁇ ⁇ is a harmonic oscillator creation operator with frequencies ⁇ ⁇ .
• an important aspect of the exemplary method of the invention is to couple the transformed and preconditioned discrete microarray output to a mathematical model for a quantum-mechanical dynamical system with specific properties.
• Specific exemplary parameters for use in calculating the Hamiltonian are those proposed by A.J. Legett et al., Reviews of Modern Physics, 59, 1, 1987 and A.O. Caldiera and A.J.Legett ⁇ wr ⁇ /s of Physics, 149, 374, 1983.
• the parameters are important only for an offline simulation of this spin Boson system on a digital computer.
• the empirical observables are then collected and used to estimate and compute spectral properties, which are actually used by the method.
• the power amplitudes ⁇ m in the mth frequency component of asymptotic state space are calculated at step 404 using
• ⁇ m ( ⁇ , ⁇ ) 4 ⁇ ⁇ P m ( ⁇ , ⁇ )
• the analytic for the external force is given by
• the parameters ⁇ , ⁇ 0 are predetermined and are design specific. Typically, values of
• the QEF is designed by matching the power spectral density (PSD) amplitude of coded mutation signature to that of the spin-boson system described above so that stochastic and deterministic time scales match and so that the time scales couple back to noise statistics and degree of asymmetry.
• PSD power spectral density
• the method employs a fully automated iterative conjugate gradient relaxation method for Spectral matching between asymmetric base system and coded mutation signature.
• the actual determination of the QEF depends on the specifics of bioelectronics substrates used for actual analysis. The method is however generalizable to all or almost all arrayed embodiments. In addition, the method is highly scalable to array dimensions (as the offline design trade-space time does not matter to computational complexity). Since the system is an overdetermined coupled system, convergence criteria and stability of relaxation method directly relates to downstream resonance effectiveness. Order Function
• the order function (OF) of ground truth is generated as follows.
• the order function (OF) is for ground truth wherein ground truth represents a state wherein a positive signal to noise ratio (SNR) is expected for hixel intensities of selected oxels.
• SNR signal to noise ratio
• the OF is calculated using an order match which implies the variance of the density function for a specific exponential family approaches 0 (within 0.000001 - 0.0001).
• the free energy for the density function is given by
• ⁇ " and ⁇ h represent state vectors.
• X a b represents the random observable in symmetric bilinear form, and ⁇ denotes the characteristic function.
• the OF derivation is based in Diado' s theory of multibranch entrainment of coupled nonlinear oscillators, wherein a number of different entrained states co-exist.
• the OF of ground truth is modulated at step 410 to yield the QEF as follows. Under controlled calibration, as stated above, maximal SNR enhancement (optimal resonance) is achieved when OF yields a single peak. It is a important design point for matching PSD of coupling spin Boson system to the synthetic QEF.
• the specific form of the QEF to be used is the generic OF shown above. So the exemplary method exploits two connotations of OF: (a) parametric form for the QEF (that is closer to the classical form) and (b) as exponential attractor for a dissipative system. The two OF's are then recoupled and convolved with the preconditioned dot spectrogram (see below).
• the QEF is represented digitally using a matrix or array having the same number of elements as the dot spectrogram to be analyzed.
• the dot spectrogram generated at step 206 is not a phase-space representation of the output of the hybridization chip, then it is desirable to convert the preconditioned dot spectrogram (generated at step 208) and the QEF into phase space to facilitate a phase space resonance interaction.
• an amplitude-based resonance interaction is performed and hence it is not necessary to convert to phase space.
• other types of resonance interactions may be employed.
• the dot spectrogram has phase space components and hence the following conversion steps are applied to the QEF at step 204 and similar steps are applied to the dot spectrogram following image preconditioning.
• a phase embedding operator, Y is applied to the hixels corresponding to the coded base mutation set such that hixel values now correspond to angles and not to intensities. These values are cyclic, with absolute magnitude of the phase image having no meaning. The relative magnitudes are more significant. So if X is a phase image and a is any constant, then X+a (mod color_map_scale) is a valid descriptor for the phase image. More importantly, the difference between a max hixel intensity and min hixel intensity is 1 and not color_map_scale. The difference between phase values at two hixels i and j is X(i)-X(j) (mod color_map_scale).
• phase embedding operator is designed such that transitive closure between any two hixels is maintained, i.e., there is an accumulated phase function ⁇ for which
• phase values of ⁇ accumulate rather than cycling back to 0.
• ⁇ can be approximated at least locally by a linear function / with deviation error function err(p) having values relatively close to zero for which
• This function / is the analog of an amplitude image having a constant intensity and err(.) is the analog of the deviation from this constant amplitude. It is assumed that err(p) never crosses a phase discontinuity and can be treated like a real-valued function. If values are computed for this deviation function err(.), then
• Each phase angle corresponds to a point on the unit circle in the coordinate lane. This is a one-to-one mapping in both direction.
• V be the mapping that takes a phase angle into the corresponding vector on the unit circle.
• I be the inverse mapping that takes a vector on the unit circle back into the corresponding phase angle.
• V(f(p)) is thus a vector valued function which has no associated phase discontinuity. V can be averaged over a region without the problems associated with directly averaging phase angles.
• N(p,m,n) of the hixel p let v be the vector which is the average of all the values of V(f(p)) in this region and let
• this line will specify a chord of the unit circle.
• the arc of the unit circle which corresponds to this chord is the region from which the "average" vector comes.
• the length of this arc is 2*arccos(
• N(p,m,n) is a symmetric mXn rectangle of hixels with p in the exact center. Since L is linear and
• phase values being averaged are all within a single phase cycle and, preferably, within a fraction of a phase cycle (e.g., 120 degrees). If the available phase image is too grainy, it may not be possible to average over a large enough rectangle. In this case, it is probably necessary that there be some larger scale linear regularity which extends over at least several consecutive phase cycles. The remaining sections presents alternatives based on this assumption.
• dh(f,p) f(i,j+l) - f(i,j) (mod)
• V ave( dv(f,p), p,m,n)
• H' ave( dh(f,p), p,m,n)
• K' for K in order to determine the function L', choose a good estimate of K' to be a phase value between -180 and +180 which minimizes the average value over the rectangle N(p,m,n) of either
• the first differences of the first differences should be approximately 0.
• the second vertical difference of a phase image f
• may be spread out over a number of hixels. This will not only make it harder to detect a significant change in d2v and d2h but also make it harder to locate the center of this change using a local maxima finding filter. It may be possible to compensate for this by averaging d2v and d2h or
• the coupling is advantageous for at least two reasons: 1) it applies to any microarray device that may provide a phase or cyclical (modulo) input as opposed to amplitude input; and 2) it is important for use with active hybridization devices which will have an element of built in control that will have a phase representation.
• the coupling is actually introduced into the exemplary method in three possible ways.
• Entrained states are employed, in part, to precisely compensate for that point. So already in the method synthetic decomposition and coupling are provided. But it is used to reject spurious candidates as opposed to "light up" more oxels with potential match. Again the step of converting to phase space may be optional depending upon the implementation and is applied, if at all, after preconditioning and/or before introducing the QEF to the resonant interaction step.
• a major limiting restriction in QSR that is avoided by the exemplary method pertains to matching the stochastic and deterministic time scales in "domain system" and the external coupling asymmetric dynamical system, since this has limited applicability to continuous data.
• phase-space representation of the resonance stimulus generated at step 204 is given by:
• ⁇ resonance stimulus (D2h(QEF MRC J) ® mod . D2v(QEF MRC ; ) for all QEF subarray elements.
• the resonance stimulus pattern is preferably represented digitally using a matrix or array having the same number of elements as the dot spectrogram to be analyzed.
• a dot spectrogram is generated at step 206 for a sample from an N by M DNA chip array wherein an element of the array is an "oxel": o(i,j).
• oxel an element of the array
• Each array cell amplitude is given by ⁇ (ij) for i: 1 to N, and j: 1 to M.
• the complimentary strand, derived from unknown sample is denoted by ⁇ (i,j) .
• the post-hybridization microarray is treated mathematically using the machinery of equations with aftereffect.
• Each hixel given by ⁇ (i,j) is represented as a cluster of dynamical systems of potentially [CB] correctly bound, [UB] unbound, [PB] partially bound and [IB] incorrectly bound.
• [CB] ⁇ (iJ) + [UB] ⁇ (iJ) + [PB] ⁇ (iJ) + [IB] ⁇ (iJ) T ⁇ (iJ) within 0.0001%.
• the fluorescence stabilization section is given by
• ⁇ l h ⁇ h (t + ⁇ ) , ⁇ ⁇ 0;
• the resulting dot spectrogram generated at step 206 is given by:
• the dot spectrogram ⁇ (i,j) is preconditioned by performing the following steps. First, at step 502, the dot spectrogram is refocused to yield a refocused dot spectrogram. Then, at step 504, a cross-correlation convolution operation is performed to yield a correlated refocused dot spectrogram. A local maxima filter ⁇ is then applied at step 506 to the correlated refocused dot spectrogram to yield a maximized dot spectrogram. The maximized dot spectrogram is re-scaled at step 508 to yield a uniformly re-scaled dot spectrogram.
• the uniformly re-scaled dot spectrogram is then further re-scaled at step 510 by amplifying local edge hixel boundaries of the uniformly re-scaled dot spectrogram to yield a globally re-scaled dot spectrogram denoted by ⁇ (i,j). Also, amplitude wanderings within the globally re-scaled dot spectrogram are estimated at step 512 for use downstream.
• DST dot spectrogram tessellation
• Purpose of DST Operator is to determine idealized ensemble boundaries for forcing downstream resonant action.
• high pass or band pass spatial filtering is implemented to enhance SNR in dot spectrogram matrix.
• Alternate methods apply a combination of Laplacian or other edge detection filters apply to enhance signal from arrays cells with a higher hybridization concentration from those of the adjacent cells.
• These SNR enhancement methods however work only with positive or zero-SNR. Since SNR in general is negative in our case (ultra- low target DNA concentrations), these methods in effect amplify noise or further blur the hixel boundaries.
• a filter ⁇ is applied to the normalized amplitude dot spectrogram matrix ⁇ , for all combinations ⁇ i+j+k where k is typically ranges from 0 to 2.
• Refocusing of the dot spectrogram at step 502 to yield a refocused dot spectrogram is performed by determining a locally averaged amplitude sub-array represented by ⁇ , then for each value of i, j and k, where k ranges from 0 to 2: determining a local standard deviation ⁇ over a (2k+l)X(2k+l) hixel neighborhood centered at (ij) and applying a filter ⁇ to the dot spectrogram ⁇ , for all combinations ⁇ i ⁇ j ⁇ k.
• ⁇ is independent of the oxel layout. From a preconditioning effectiveness standpoint, a best case design corresponds to a completely random oxel layout in terms of O -value for adjacent oxels. The worst case corresponds to O -value separation of 1 among adjacent oxels.
• the oxel, o(ij) z , centered at (ij) comprises of complementary oligonucleotides, corresponding to a mutation of interest over the set Z. Loci, r, for averaging amplitude ranges from +5 oxels to ⁇ k oxels depending on ( O -value mod 4 k ) separation. ⁇ f then
• composite loci, r' is bounded by a rectangle whose top left hand and bottom right hand corner coordinates are given by [i m ⁇ n -k-l, j max +k+l] and [i max +k+l, j m ⁇ n -k-l] where i mm , i max> i n ana ⁇ J max correspond to the ordinate and abscissa for the oxel mutations with overlapping loci.
• the preceding expression captures local ensemble deviation from dot spectrogram average.
• the design parameter K is computed offline based on specific bio-molecular signatures of interest.
• the suffix / denotes iterative index for a cross-correlation convolution operation to be applied after global refocusing.
• a uniform dot spectrogram rescaling can be achieved by applying ⁇ .
• ⁇ (Ij) where ⁇ can be either a constant or a functional. This operator selectively enhances those hixels and ensemble boundaries whose intensity exceeds the local average by more than K prespecified standard deviations.
• K is a predetermined constant computed offline based on microarray fluorescence or chemiluminiscence sensitivity.
• the suffix / denotes an iterative index for a cross-correlation convolution operation to be applied after global refocusing.
• the cross-correlation convolution operation of step 504 to yield a correlated refocused dot spectrogram is performed by convolving the refocused dot spectrogram with an apriori-chosen restricted field obtained using a strongly dissipative dynamical system.
• Dissipative dynamical systems are those which define a forward regularizing flow in an adequate phase space containing an absorbing set.
• An absorbing set is a bounded set that attracts all bounded solutions in finite time at an exponential rate. Since we exploit a strongly dissipative system, the absorbing set is required to be unique compact set that is both positively and negatively invariant under the flow, it attracts all flows.
• the post-hybridization hixel denotes a projection of the flow field that is absorbed by the coupling system. Since we couple with an absorbing set, this stage yields significant SNR enhancement.
• the convolving field can be constructed using Kuramoto-Sivashinsky equation, 2D Navier-Stokes equation or some forms of Reaction-Diffusion equations.
• the dot spectrogram subarray around the oxel detecting mutation of interest can be cross-correlated with any dissipative dynamical system. In summary this step exploits proven classes of mathematical system or exponential attractors.
• exponential attractor is an exponentially contracting compact set with finite fractal dimension that is invariant under the forward flow.
• A -P H ⁇ is the Stokes operator
• B(u,u) stands for the non-linear term (u . ⁇ u) projected to the underlying Hubert space H
• f is the volume force projected to the same Hubert space
• v is the viscosity term.
• Galerkin Approximation can be used to approximate the exponential attractor for the system
• ⁇ denotes the positive eigenvalue of A.
• the rate of convergence of this system can be computed as well.
• the exponential attractor is then coupled with the post-hybridization dot spectrogram subarray
• the exponential attractor is discretized over the grid that corresponds to the refocused amplitude subarray associated with a mutation and estimated above.
• the actual convolution of the two systems is then given by
• This process is computed for each oxel associated with the loci of a specific mutation of interest.
• Step 506 for applying a local maxima filter ⁇ at to the correlated refocused dot spectrogram to yield a maximized dot spectrogram is then implemented as follows.
• the local maxima filter is defined by
• the maximized dot spectrogram is rescaled at step 508 to yield a uniformly re-scaled dot spectrogram by applying an operator .
• the uniformly re-scaled dot spectrogram is itself then rescaled at step 410 by amplifying local edge hixel boundaries of the uniformly re-scaled dot spectrogram to yield a globally re-scaled dot spectrogram. This is achieved by 1 ) determining the zero mean amplitude for the uniformly rescaled dot spectrogram; 2) applying a logarithmic rescaling function p around the zero mean amplitude; and 3) merging the local maxima into a single local maximum halfway in between.
• the logarithmic rescaling function is generated by generating an expansion sequence of nonnegative numbers and by generating an expanded dot spectrogram tessellation for ⁇ .
• the expansion sequence is generated as follows:
• the expanded dot spectrogram tessellation for ⁇ (which is represented by DST( ⁇ )) is generated using:
• D( ⁇ j, ⁇ k ) is a shortest possible discretized fluorescence amplitude separation between a pair of hixels ⁇ i5 ⁇ k wherein ⁇ k is a local maximum.
• the local maxima are merged into a single local maximum half way in between for downstream hixel-to-ensemble and ensemble-to-ensemble operations on hixel clusters using the sequence
• ⁇ k be a local maxima image for some other dot spectrogram realization ⁇ (/, ) , .
• D( ⁇ ; , ⁇ k ) to be the shortest distance possible from hixel ⁇ j to some hixel ⁇ k which is a local maximum.
• a Dirichlet tessellation operator or a Delaunay triangulation operator are then applied to perform gradient refocusing rather than steps 502 - 510.
• Amplitude wanderings are estimated at step 412 within the enhanced dot spectrogram.
• the estimate is performed by applying a Palm Distribution operator to the globally re-scaled dot spectrogram to capture amplitude wanderings and transitions at element, neighboring pair and local ensemble levels.
• the application of the Palm Distribution operator generates bounds that are used to accommodate degradation of hybridization over time.
• the estimate exploits the use of generator functions to capture stochastic variability in hybridization binding efficacy and draws upon results in stochastic integral geometry and geometric probability theory.
• Geometric measures are constructed to estimate and bound the amplitude wanderings to facilitate detection.
• MRC- mutation-recognizer centered
• Such a representation uniqueness facilitates the rapid decimation of the search space. It is implemented by instantiating a filter constructed using measure-theoretic arguments.
• the transformation under consideration has its theoretical basis in the Palm Distribution Theory for point processes in Euclidean spaces, as well as in a new treatment in the problem of probabilistic description of MRC-hixel dispersion generated by a geometrical processes. Latter is reduced to a calculation of intensities of point processes. Recall that a point process in some product space E X F is a collection of random realizations of that space represented as ⁇ (esammlung Q,
• Palm distribution, ⁇ of a translation (T n ) invariant, finite intensity point process in 9T is defined to the conditional distribution of the process. Its importance is rooted in the fact that it provides a complete probabilistic description of a geometrical process.
• Palm distribution can be expressed in terms of a Lebesgue factorization of the form
• the oligonucleotide density per oxel p m(l ) , PCR amplification protocol ( ⁇ m ), fluorescence binding efficiency ( ⁇ m ) and imaging performance (w m ) provide the continuous probability density function for amplitude wandering in the m-th MRC-hixel of interest. Let this distribution be given by (p m(I ) , ⁇ m , ⁇ m , a ⁇ m ) .
• the method requires a preset binding dispersion limit to be provided to compute A.
• ⁇ , and ⁇ 2 represent the normalized hybridization dispersion limits. These number are empirically plugged in. We choose 0.1 and 0.7 respectively to signify loss of 10% - 70% hybridization.
• ⁇ denotes the distribution of known point process. We use the form 1 /( 1 +exp( f rigid)) to represent it.
• the final preconditioned (or enhanced) dot spectrogram generated by the step of FIG. 5 is represented by:
• step 210 the resonant interaction between the QEF and the preconditioned dot spectrogram is performed until a pre-selected degree of convergence is
• V(u) is the actual precondition, refocused MRC-hixel subarray. So a stable equilibrium state (microarray) is transformed and modulated with the QEF, i.e., mathematically destabilize it to achieve a nonlinear resonance point.
• u fluorescence decoherence timescale(s).
• State variable u corresponds to nonstationary Markov random field (NS-MRF).
• Y(t) Asin (wt + 1) where t is a small random phase factor.
• A is a gain function (control parameter that influences convergence rates),
• t denotes integration timestep.
• QN(t) corresponds to the log (PSD maxima) of the ground truth order function summed over all regions of interest. More precisely, QN(t) is reverberation projection (at some instant t prior to resonant convergence) for the coupled Ofs discussed above in relation to step 204.
• Palm generators are used as additive correction terms to the potential gradient to compensate for uncertainty and post hybridization decay.
• the actual dynamics is given by V(u, II, ⁇ ).
• a new QEF (corresponding to a new mutation of interest) is selected and the method reinitializes to the original DS as computed in step
• This technique referred to herein as a "software programmable QEF flush control methodology", of reinitializing the dot spectrogram and using a new QEF leads to a new cycle through a signal sorter for rapid single- / multi-point gene/mutation sorting. It is amenable to RF, electronic as well as optoelectronic QEF loading to microelectronics computing backplane, microarray readout and analysis backplane itself optoelectronically or electronically bonded to bioelectronic substrate. Furthermore the method can be implemented on offline miniature custom VLSI / palmtop / desktop setup. The QEF is also implementable in electronic, optoelectronic, bionic and ionic COTS/custom device physics.
• a chip fabricated to implement the exemplary method can work in two ways: a) seek all mutations of interest simultaneously; or b) seek all mutations of interest serially.
• An advantage of (a) over (b) is computational speedup.
• An advantage of (b) over (a) is serendipity, i.e., in method (a) only those mutations are resonantly amplified that are detection candidates. Everything else is likely suppressed or decimated.
• multiple resonant outputs can be accepted. By accepting multiple peaks for the OF the method can actually accept 2 nd order, 3 rd order entrained states, where order implies hamming distance to the mutation of interest in terms of base pair labels and locations. This can be used to theoretically accept entire families of derived mutations.
• the resonance interaction is performed digitally by applying a matrix representative of the resonance equation to a matrix representative of the resonance stimulus in combination with a matrix representative of the dot spectrogram.
• the final result of step 210 is a set of hixel locations wherein resonance has occurred identified by row and column number (i j).
• the hixel addresses (k,l) of those locations are mapped at step 212 into the oligonucleotide table mentioned above (which contains the oligonucleotide sequences associated with hixel locations) to thereby identify the mutations, if any, present in the sample being analyzed.
• This is a simple table look-up resulting in a direct readout of the mutations. For a custom POC diagnostic sensor only those hixels which relate to mutations or expressed products of interest are stored in the table.
• the mutations identified at step 212 are then mapped into the mutations table mentioned above (which contains the diseases associated with the mutations) to thereby identify the diseases, if any, present in the sample being analyzed. This is also a simple table look-up resulting in a direct readout. Again, for a custom POC diagnostic sensor only those mutations which relate to diseases of interest are stored in the table.
• the resonant output interaction is interpreted to yield a set of confirmed mutations as follows:
• MRC-hixeli z o(h(k,l)) where o is some hashing function or table look-up.
• the step will readout the oligonucleotide sequence from a table that has encoded the microarray. Note that this mapping also operates to map the confirmed mutations to known diseases associated with the pre- selected set of mutations of interest to identify diseases, if any, indicated by the DNA sample (step 214). Note also that no probabilistic inferencing, exploitation of learning or nonlinear mapping is required to interpret the resonance output. Rather interpretation is rendered very straightforward thereby requiring only a low-cost hardware implementation with simple software to implement steps 212 and 214. So an important aspect of the exemplary method is to accurately and robustly detect specific oligonucleotide sequences in the target sample. Subsequent association to understood genomics pathways is trivialized.
• step 214 the diagnosis generated by step 214 is confirmed at step 216 by taking the identified diseases and solving in reverse for the associated QEF and then comparing the associated QEF's with ones expected for the mutations associated with the identified diseases to verify correspondence and, if correspondence is not found, then a new set of mutations of interest are selected and all steps repeated.
• this step maps detected mutations and expressed genes to a diagnostic assessment.
• This is a probabilistic or deterministic step, depending upon the genomics of the specific disease. It is represented as
• Step 216 is a double check mechanism adopted to confirm multi-factorial diseases where the biochip encodes complex genomics.
• This implementation which may also be referred to as a "mixed-mode phase shifted mode" is particularly effective for automatically extracting an entire class of mutations that may be manifested in a hybridized element.
• the mixed-mode provides polymorphism in induced couplings for QEF design which delivers repeatability on analysis whereby mutation signatures of interest are simultaneously coupled to many base "dynamical systems" with a single phase-embedding operator. Other resonance coupling interactions may be exploited as well.
• couplings other than phase-based, are "additive coupling mode” which provides further SNR enhancement and a “shunted input multiplicative coupling mode” which amplifies noise-to-noise couplings and leads to derivation of better readout threshold for diagnostics decision making. Also, a combination of different resonance interactions can be exploited.

Landscapes

• Engineering & Computer Science (AREA)
• Health & Medical Sciences (AREA)
• Life Sciences & Earth Sciences (AREA)
• Physics & Mathematics (AREA)
• Medical Informatics (AREA)
• Chemical & Material Sciences (AREA)
• Biophysics (AREA)
• Bioinformatics & Cheminformatics (AREA)
• Nanotechnology (AREA)
• Biotechnology (AREA)
• Molecular Biology (AREA)
• General Health & Medical Sciences (AREA)
• Theoretical Computer Science (AREA)
• Spectroscopy & Molecular Physics (AREA)
• Genetics & Genomics (AREA)
• Evolutionary Biology (AREA)
• Bioinformatics & Computational Biology (AREA)
• Crystallography & Structural Chemistry (AREA)
• Pharmacology & Pharmacy (AREA)
• Epidemiology (AREA)
• General Engineering & Computer Science (AREA)
• Signal Processing (AREA)
• Artificial Intelligence (AREA)
• Bioethics (AREA)
• Computer Vision & Pattern Recognition (AREA)
• Data Mining & Analysis (AREA)
• Databases & Information Systems (AREA)
• Medicinal Chemistry (AREA)
• Evolutionary Computation (AREA)
• Public Health (AREA)
• Software Systems (AREA)
• Mathematical Physics (AREA)
• Measuring Or Testing Involving Enzymes Or Micro-Organisms (AREA)
• Apparatus Associated With Microorganisms And Enzymes (AREA)
• Investigating Or Analysing Biological Materials (AREA)
• Investigating Or Analyzing Materials By The Use Of Electric Means (AREA)
• Management, Administration, Business Operations System, And Electronic Commerce (AREA)
• Image Processing (AREA)

Priority Applications (1)

Applications Claiming Priority (3)

Related Child Applications (1)

Publications (2)

Family

ID=22961710

Family Applications (2)

Family Applications Before (1)

Country Status (6)

Cited By (1)

Families Citing this family (75)

Family Cites Families (66)

• 1999
  • 1999-02-22 US US09/253,789 patent/US6136541A/en not_active Expired - Lifetime
• 2000
  • 2000-02-17 AU AU55859/00A patent/AU5585900A/en not_active Abandoned
  • 2000-02-17 EP EP03028721A patent/EP1406201A3/fr not_active Withdrawn
  • 2000-02-17 DE DE60015075T patent/DE60015075T2/de not_active Expired - Lifetime
  • 2000-02-17 WO PCT/US2000/004076 patent/WO2000052625A2/fr active IP Right Grant
  • 2000-02-17 US US09/913,132 patent/US6671625B1/en not_active Expired - Lifetime
  • 2000-02-17 EP EP00941101A patent/EP1145181B1/fr not_active Expired - Lifetime
  • 2000-02-17 AT AT00941101T patent/ATE280415T1/de not_active IP Right Cessation
• 2003
  • 2003-08-12 US US10/639,608 patent/US6963806B2/en not_active Expired - Fee Related

Non-Patent Citations (1)

Also Published As

Similar Documents

Legal Events